International Journal of Innovations in Science & Technology
Not a member yet
813 research outputs found
Sort by
Optimization of MPPT in PV Systems Using Machine Learning Under Partial Shading Conditions
Full text linkPhotovoltaic (PV) systems are an important solution to the increasing global demand for electricity and the declining availability of fossil fuels. However, under Partial Shading Conditions (PSC), the Power-Voltage (P-V) curve can have multiple local peaks, which leads to significant power losses and makes it harder to find the true Maximum PowerPoint (MPP). Traditional algorithms like Perturb and Observe (P&O) and Incremental Conductance (INC) often mistake these local peaks for the global ones, making it difficult to accurately track the Global Maximum PowerPoint (GMPP) during shading. To overcome this issue, Machine Learning (ML)-based Maximum Power Point Tracking (MPPT) methods are explored as a data-driven alternative. These aim to improve accuracy and reduce energy loss in PV systems affected by shading. The study evaluates several ML techniques—Artificial Neural Networks (ANN), Support Vector Machines (SVM), Decision Trees (DT), Random Forests (RF), and Weighted K-Nearest Neighbours (WK-NN) using both synthetic and real-world weather data from Johannesburg, South Africa. To test their effectiveness, the models are simulated and implemented on a hardware-based PV system. Results show that ML-based MPPT methods significantly enhance tracking performance and reliability. For example, SVM achieves an efficiency of 96.76% under normal conditions and 83.66% during heavy shading, while ANN reaches 99.58% efficiency in stable sunlight. RF and WK-NN also maintain over 95% efficiency in changing conditions due to their adaptability. Despite the promising results, some challenges remain. These include computational complexity, real-time deployment limitations, and the ability of models to generalize under varying sunlight levels. Still, this study demonstrates that AI-powered MPPT systems can greatly improve energy management and grid stability in next-generation solar technologies. Future research should focus on deep learning-based MPPT, hardware-efficient AI models, and real-time optimization to reduce processing demands and improve scalability in embedded MPPT controllers
An Automated Approach for Enhancing Efficiency and Transparency in Student Selection Process for Public Sector General Universities
Full text linkThe admission process in public sector universities in Pakistan faces challenges, including a large volume of applications, complex eligibility criteria, and the need for equitable seat allocation across various quotas. Many public sector universities still rely on manual or semi-automated admission systems, which result in inefficiencies related to time and transparency. Furthermore, these systems are vulnerable to errors in the seat allocation process due to human involvement at certain stages.
To address these issues, this paper proposes a fully automated admission system for public sector general universities. The system is developed and implemented at the University of Sindh, Jamshoro, one of Pakistan’s oldest and largest public sector universities. Following the successful implementation of the system, a performance evaluation and comparative analysis are conducted to assess its effectiveness and confirm its feasibility for all public sector general universities in Pakistan.
Additionally, a usability study is carried out to ensure the system\u27s flexibility and ease of use from the user\u27s perspective. The results from the usability study and comparison indicate that the proposed system outperforms existing systems in terms of flexibility, reliability, efficiency, and transparency
Assessing the IoT Acceptance at Public Sector Universities of Sindh, Pakistan
Full text linkThe Internet of Things (IoT) technology blends the real world and digital life, ensuring seamless integration for accomplishing tasks to make life easier. Analysts often get lost in the deep technical details of IoT, but there is a lack of focus on student acceptance and willingness to adopt these technologies. Concentrating on the factors that drive the adoption of IoT technologies. This study employs a quantitative approach to investigate the deep interrelations and interactions in the process with the Unified Theory of Technology Acceptance (UTAUT2) and other factors like IoT Skills, Trust, and Personal Innovativeness. Through an explanatory survey method, data was collected from 389 students across 5 public universities in Sindh, Pakistan, to assess the level of acceptance of IoT technologies in universities among the students. This study produces existing literature by expanding the UTAUT2 model to incorporate novel elements relevant to the acceptability and application of IoT in developing nations. It provides important recommendations for policymakers and university stakeholders. The results highlight the need for improving IoT infrastructure, incorporating central IoT courses in academic offerings, and developing an enabling environment for successful technology adoption. The evidence presents inadequate proper IoT infrastructure and supporting environment in institutions. In addition, the adoption of IoT among students is evidenced by the study field instead of by the professional need for IoT
Optimizing Economic Load Dispatch Using a Hybrid PSO-SA Algorithm: A Novel Approach
Full text linkEconomic Load Dispatch (ELD) is a crucial power system optimization task. It aims to minimize the total cost of electricity generation by strategically allocating power output among available generating units to meet the system\u27s demand while respecting operational limits. This paper investigates how soft computing methods can improve the effectiveness of Electronic Logging Device (ELD) solutions. Specifically, Particle Swarm Optimization (PSO) and Simulated Annealing (SA) algorithms are employed to minimize generation costs for a power system comprising three generating units. The optimization process considers loss coefficients, generation limits, and a predefined cost function. Initially, PSO is used to determine near-optimal solutions, which are further refined using SA to avoid local minima. A hybrid PSO-SA method integrates the global exploration of Particle Swarm Optimization (PSO) with the local refinement of Simulated Annealing (SA) to enhance convergence and solution quality. 1 This approach was implemented in MATLAB and validated through a case study. Simulation results demonstrate that the hybrid method consistently yields high-quality solutions with reduced computational effort, proving its robustness and reliability for solving ELD problems. Combining metaheuristic algorithms shows promise for real-world power system optimization
Distributed Denial of Service (DDOS) Attacks Technique to Interruption the System\u27s Service and Identification
Full text linkDistributed Denial of Service (DDoS) attacks remain to present significant threats to network stability and security by flooding systеms with malicious traffic intеndеd to intеrrupt legitimate sеrvicеs. This dissertation looks at numеrous DDoS assault tactics and assеssеs thеir detection and mitigation using Snort and an opеn sourcе nеtwork intrusion detection system (NIDS). To adequately invеstigatе thеsе assaults and a thorough networks architecture was created and simulatеd with GNS3 and which includеd many VMwarе virtual machines to imitate a realistic network еnvironmеnt. Thе research investigates a variеty of DDoS attack tactics and such as volumetric assaults that flood the network with excessive data, protocol attacks that exploit vulnerabilities in network protocols, and software layer attacks that specifically target certain apps or services. The networks architecture gеnеratеd by GNS3 еnablеd thе controlled deployment of diffеrеnt attack vectors and offering insights on thеir influеncе on nеtwork performance and security. Snort was usеd to dеtеct and analyze thеsе assaults and taking usе of its rulе based detection capabilities to discover patterns and abnormalities associated with DDoS activity. Thе study assesses Snort\u27s еfficacy in detecting and rеacting to various DDoS attack signatures and with a focus on its rеal timе analysis of\u27 alerting systеms. Thе findings show Snort\u27s strеngths and limits in controlling various forms of DDoS assaults and offеring usеful insights into its rolе in improving nеtwork sеcurity. Furthermore, and thе study еmphasizеs thе nееd of a strong nеtwork architеcturе and ongoing monitoring in protecting against merging thrеats. Thе research presented hеrе contributes to our undеrstanding of DDoS attack dеtеction and thе actual implеmеntation of Snort in simulated network settings and including techniques for strengthening community resilience against attacks
Enhanced Skin Cancer Classification with MobileNetV3 and Morphological Preprocessing: A Deep Learning-Based Extension
Full text linkSkin cancer detection continues to pose challenges due to the visual similarity between the types of lesions and the limitations of traditional diagnostic methods. This study presents an extended and improved skin lesion classification framework that combines transfer learning with MobileNetV3 and enhanced preprocessing using mathematical morphological techniques. These preprocessing methods refine lesion boundaries and suppress irrelevant structures in dermoscopic images, thereby improving feature discrimination during training. The refined framework is evaluated using the ISIC dataset and achieves a notable classification accuracy of 89%, showing superior performance compared to baseline models. This extension also examines the generalizability and suitability of the model for deployment in low-resource mobile settings. The results validate the effectiveness of lightweight architectures paired with morphological enhancements, providing a reliable and scalable solution for early skin cancer screening and clinical support
Hybrid Intrusion Detection System Based on Optimal Feature Selection and Evolutionary Algorithm for Wired Networks
Full text linkThe field of cybersecurity encounters ongoing difficulties in identifying and preventing attacks in networks, and the pervasive threat of cyberattacks demands continual advancements in intrusion detection systems (IDS) to safeguard network integrity. Traditional intrusion detection systems face the challenge of class imbalance. Addressing the formidable challenges posed by class imbalance and high-dimensional data, this research proposes a novel hybrid IDS approach. Leveraging (ACO), the algorithm navigates complex datasets to identify salient features, effectively mitigating the complexities associated with high-dimensional data. Subsequently, a Weighted Stacking Classifier amalgamates the strengths of Random Forest, AdaBoost, and Gradient Boosting classifiers, fortifying the system’s ability to handle class imbalance robustly. By strategically enhancing the importance of base classifiers with favourable training outcomes and diminishing the influence of those yielding inferior results, the hybrid IDS endeavors to optimize classification efficacy. The experimentation, conducted exclusively on the dataset named NSL-KDD, demonstrates the efficacy of the proposed model, yielding remarkable results. With a 90.13% Accuracy, 88.87% precision, 91.23% Recall, and 87.33% F1-score, the hybrid IDS exhibits superior performance in detecting malicious activity. The findings underscore the viability of the proposed hybrid IDS as a potent tool in the ongoing battle against cyber threats, positioning it for real-world deployment across diverse networks
Enhancing Non-Player Characters (NPC) Behaviour in Video Games Using Reinforcement Learning
Full text linkNPCs enrich the immersive experience of a video game, and traditionally exist along purely rule- or script-based paradigms, denying adaptability or intelligent decision-making very often. The research integrates RL into the NPC behaviour to allow for the more realistic, dynamic interactions and responsive behaviour that today\u27s gaming environments require. We will review state-of-the-art RL algorithms and validate improvements implemented in our own RL model within a sandbox game environment into NPC decision-making and player engagement. According to our results, RL makes NPCs adaptive, tactically deep, and realistic while the classical ones fail. The study provides rigorous methodology and analysis to demonstrate the feasibility and advantages of using RL for the design of a new generation of games
Advancements in Automatic Text Summarization using Natural Language Processing
Full text linkWith the rapid expansion of data across various domains, the need for automated text summarization has become increasingly crucial. Given the overwhelming volume of textual and numerical data, effective summarization techniques are required to extract key information while preserving content integrity. Text summarization has been a subject of research for decades, with various approaches developed using natural language processing (NLP) and a combination of different algorithms. This paper is an SLR-type essay presenting the existing text summarization techniques and their evaluation. It covers the basic concepts behind extractive and abstractive summarization and how deep learning models could serve as a boost in the performance of summarization. The study goes on to investigate the present use of text summarization in different areas and looks into the various methodologies applied in this area. A total of twenty-four carefully selected research articles were being analyzed to identify key trends, challenges, and limitations regarding text summarization techniques. The paper further discusses the existing literature and proposes a number of open research challenges with insight concerning possible future directions in text summarization
AI-Sentinel: A Novel AI-Powered Intrusion Detection Approach Against Cyber Threats for In-Vehicular Communication Systems
Full text linkThe emergence of revolutionizing technologies such as Artificial Intelligence and the Internet of Things, and their integration into the automotive industry has brought innovations and made the lives of common people easier and complacent. Leveraging the advanced intelligent services provided by the connected and autonomous vehicles the driving experience is much more convenient and effortless. The CAN (Controller Area Network) protocol is the most commonly deployed protocol in in-vehicular communications in the ICVs (intelligent connected vehicles) environment due to its efficiency and speed. However, it lacks basic security mechanisms like encryption and authentication making it vulnerable to various cyber threats. In this article, we have presented a novel, robust, cutting-edge AI-based Intrusion detection system for detecting various seen and unseen cyber-attacks in in-vehicular networks to ensure security. Two main models deployed in the proposed framework are RNN for dealing with temporal dependencies in the CAN traffic and LightGBM for efficient feature extraction. The experimental results show that the hybrid of these two models performs better in terms of various evaluation metrics, with its accuracy being 94% in classifying the CAN traffic into normal and different attack classes. A comparison with the existing state-of-the-art approaches shows that our proposed approach is more robust and secure, with it being deployed in a Federated Learning FL environment